thek0tyara/llama-cpp-turboquant
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions 4

cuda : fix nkvo, offload and cuda graph node properties matching (#19165)

Browse source 
* cuda : fix nkvo

* cont : more robust cuda graph node property matching

* cont : restore pre-leafs implementation

* cont : comments + static_assert
This commit is contained in:
Georgi Gerganov 2026-01-29 18:45:30 +02:00 • committed by GitHub
parent 7b7ae857f6
commit 4fdbc1e4db
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
4 changed files with 59 additions and 32 deletions
Download patch file Download diff file Expand all files Collapse all files

5
src/llama-graph.cpp
View file

@ -1630,11 +1630,6 @@ ggml_tensor * llm_graph_context::build_attn_mha(
hparams.attn_soft_cap ? hparams.f_attn_logit_softcapping : 0.0f);
cb(cur, LLAMA_TENSOR_NAME_FATTN, il);
if (!cparams.offload_kqv) {
// all nodes between the KV store and the attention output are run on the CPU
ggml_backend_sched_set_tensor_backend(sched, cur, backend_cpu);
}
ggml_flash_attn_ext_add_sinks(cur, sinks);
ggml_flash_attn_ext_set_prec (cur, GGML_PREC_F32);